Windows 2003 Network Load Balancing Question

[qshiva]

Dear Group,

I have configured Windows 2003 Network Load Balancing for my IIS application. One Cluster IP address and two physical nodes. In each node there is one Network adapter. The NLB cluster configured with multi cast and filtering is set to one. NLB is configured for port 80. The Cluster IP dares has also a hostname configured in DNS.

I noticed when I stop IIS on one node my IIS application works fine.

If I disable a Network Adapter on one node my IIS application doesn't understand it always. Meaning sometimes its running correctly sometimes not. I get a classic error message that my IIS is not reachable.

It looks like NLB sends my IIS requests to the wrong Node because NLB does not understand one NIC is disabled. The NLB cant negotiate sufficient. I'm aware of the heartbeat between the two NLB adapters. It normally takes approximately 6,7 seconds before NLB knows one machine is down.

The question is does NLB understand if I'm disabling a Network Adapter ?

Is there somebody with the same experience with NLB ?

Thanks in advance.

Best Regards,

Shiva

[Mr Snrub]

What is the affinity set to?

Is the dedicated IP address the primary one in each node, and the common virtual address added as a secondary? (This is how you should have it.)

NLBS is not application-aware, so stopping the IIS service will make no difference to the cluster membership and you should get client connections still being directed to each node as they are both working correctly.

Disabling the NIC should cause a recovergence of the cluster and the one node left should handle 100% of the load.

It is vital that you use a layer 2 switch, or a layer 2 VLAN if it is a layer 3 switch.

What does the command "wlbs query" report when you run it on either node when you are doing your tests?

Almost every NLB issue I have dealt with has been a switch/network problem.

[qshiva]

Dear Mr Snrub,

Thanks for your clear reply.

Affinity is set to none.

What do you mean with the following ?

>It is vital that you use a layer 2 switch, or a layer 2 VLAN if it is a layer 3 switch.

>Almost every NLB issue I have dealt with has been a switch/network problem.

You mean the problem is related to the network and not to the nodes.

I didnt approach the problem this way.

Anyway

I'm honder procent sure its a cisco device. I'm not sure but I think its a swich. Only I'm not sure which type. I will check this

I will run the "wlbs query" Iwant aware of this utility at that time.

Hope to hear from you soon. Its hard to find someone with experience on this topic.

Thanks a lot.

Best Regards,

Shiva

[Mr Snrub]

What do you mean with the following ?

>It is vital that you use a layer 2 switch, or a layer 2 VLAN if it is a layer 3 switch.

>Almost every NLB issue I have dealt with has been a switch/network problem.

You mean the problem is related to the network and not to the nodes.

The NLBS component is so very simple it doesn't have much to "go wrong" - when networking devices try to be too smart they cause problems, though.

In essence, switches that try to route at the IP level instead of the MAC level can associate an IP address (a virtual one in this case) with a specific port that they "know" the device is on.

Cisco, in particular, do not agree with Microsoft on the implementation of virtual IP load balancing and so their devices often cause problems.

Anyway I'm honder procent sure its a cisco device. I'm not sure but I think its a swich. Only I'm not sure which type. I will check this

I will run the "wlbs query" Iwant aware of this utility at that time.

Hope to hear from you soon. Its hard to find someone with experience on this topic.

I have a bit of an advantage

Your easiest way to prove it's a networking infrastructure problem and not with NLBS or IIS itself is to plug both servers into a hub or a simple switch which is then uplinked to the Cisco.

I've had this problem myself with a Cisco Catalyst 6500.